Which decision will this data change? We need an answer before starting the analysis. Without understanding which period the data represents and how it measures a given concept, an analysis may amount to little more than a persuasive narrative. AI can produce that narrative quickly, but it does not make the wrong question right.
When looking for the right question, I start with the strategic context: Who owns the decision, which group will be affected, and what change are we looking for over what period? Writing down the decision question, the affected group, and the expected outcome clarifies why we are examining the data.
We can then assess the data source, sample, missing records, changing definitions, and potential biases. The data's origins and representativeness are also part of this review. AI can help generate hypotheses, create a data dictionary, look for anomalies, and develop alternative explanations. Distinguishing correlation from causation, interpreting a measure's business meaning, and defining the limits of a decision depend on human data literacy.
I do not expect a data culture to develop if it is confined to tool training. Building a shared language around measurement, evidence, uncertainty, and causality helps employees discuss openly which questions they are asking and why. Asking AI for its assumptions, uncertainties, alternative explanations, and the additional evidence needed alongside its answer can put this language into everyday practice. This also makes answers easier to check and use at work.
Ask yourself
- Which decision will this data change?
- Is what we are measuring the same as what we want to understand?
- What missing or biased data could reverse the result?
